Context Has Temporal Strata — Authority Disagreement, Minimal Data Contracts, Friction Governance, Annales Historiography, and the Mobius Reflective Context Governor

Long-context AI systems increasingly fail not because they lack context, but because they receive too much undifferentiated context: durable constraints, active decisions, stale specifications, temporary tool errors, copied hallucinations, secrets, and user preferences all enter the window as flat text, and the model is asked to infer authority, freshness, provenance, and safety from position and wording alone. This paper names that failure mode contextual flattening (with the subtype temporal flattening) and proposes the Mobius Reflective Context Governor (RCGov) as a semantic-filtration and context-governance layer, under the axiom InjectContext_t ⇒ ContextReady_t. The paper is explicit about its confidence levels. The robust core: (1) context should pass non-compensatory gates before it is scored or injected; (2) authority proposals must be separated from authority commitments; (3) when human authority labels have low inter-rater agreement, the system should surface disagreement (kappa-first) rather than pretend classification has solved authority; and (4) disagreement visibility must itself be governed through friction governance (Disagreement Fatigue Index, Override Risk Score, elastic thresholds, reasoned overrides) so alert fatigue does not induce bypass. The experimental bet, kept separate with a pre-stated demotion criterion: Annales-derived temporal strata (after Braudel) may improve attention-aware pack placement. The v0.7 text is synchronized with a minimal implementation contract (five authority states, four temporal strata, ten segment roles) and specifies the governance artifacts a first implementation should emit: Clean Context Pack, Non-Injection Report, CONFLICT_MAP, and an authority review queue. Deposit 1 of 2; the companion record bundles the controlled N=120 RAW-vs-CLEAN evaluation with the frozen Specification v0.4 and Minimal Data Contract v0.1. AI co-observer: project AI systems assisted drafting; this deposit edition was prepared with Claude Fable 5 (Anthropic), working method only — the registered author is the human author alone.